Luke Longren: Automatic Segmentation of Elephant Trunk Muscles

BCCN Berlin / Technische Universität Berlin

Abstract

Muscles provide mammals a mechanism for movement. A particularly interesting system of muscles is the elephant's trunk. Forming a muscular hydrostat, the trunk is able to manipulate objects with great precision. While the morphology of the trunk has been studied in detail over the past two centuries, a precise count for the number of individual muscles the trunk contains has not yet been determined; previous estimates up to 150,000 have been made. A justifiable muscle count enables a quantitative comparison to the number of motor neurons that innervate the trunk. In the present thesis, an automatic method of muscle segmentation is investigated. Computed tomography is used to obtain volume images of the trunk, upon which the method implements a basic deformable model with a vector field as the main external force. To determine the external force, a vector field convolution from the edge map is taken. For testing the deformable model's implementation, a variety of edge map sources are used; geometries range from the surfaces of simple shapes to Sobel operator edge detection. To provide improved convergence, a more informed initialization process is investigated as well. Results showed success of the model segmenting simple shapes and true edge maps. However, limitations of the current model are present due to physical complexities of the trunk and the subsequent inability to generate an adequate edge map in an entirely automatic method. Future work to establish a semi-automatic segmentation method using a minimal amount of manual tracing for initialization is planned, potentially leading to individual muscle segmentations for an entire transverse section of the trunk.

 

Additional Information

Master Thesis Defense

 

Organized by

Prof. Michael Brecht   & Dr. Daniel Baum     / Lisa Velenosi